Key points
- Anthropic has onboarded six METR employees, giving them company hardware, office space and internal tools roughly on par with its own risk assessment teams.
- Anthropic is not paying METR, which does not accept donations from AI companies; METR can publish its analysis without Anthropic veto.
- Anthropic can require redactions in set categories such as customer data, attorney-client privileged details and model architecture information.
- OpenAI has not yet laid out plans for evaluators, though Sam Altman called Dario Amodei's September proposal a good idea.
Why it matters for marketers: Marketers who build on frontier models get an early look at how vendors may be held to outside scrutiny, which could shape vendor risk reviews and procurement questions.
Anthropic has given the nonprofit METR what it calls employee-like access to assess the risks of its models. Six METR staff now work with company hardware, office space and internal tools that Anthropic says are roughly equal to what its own risk assessment teams use. The company says it has encouraged employees to share information with them as they would with a colleague.
The arrangement follows a proposal from chief executive Dario Amodei in September to embed external auditors at leading labs, which OpenAI's Sam Altman and Elon Musk backed. Anthropic is not paying METR, which is funded by individuals, philanthropies and government bodies such as the UK's AI Security Institute. METR can publish its findings without a veto, though Anthropic can require redactions in certain categories. Critics argue such setups could create conflicts of interest and that stronger government regulation is needed.
For marketers and content teams, this is mainly a vendor-trust signal. Many teams now run agents and content workflows on top of frontier models, and questions about model safety increasingly show up in procurement and legal reviews. Published third-party analysis could give those reviewers evidence beyond a vendor's own claims. The timing also matters: the Financial Times reports Anthropic is under pressure to reassure investors about safety ahead of an expected IPO this year.
Watch whether OpenAI and other labs announce comparable access, and what METR actually publishes. Teams evaluating AI vendors may want to ask each one whether independent evaluators have access and what limits apply to disclosure.
Sources
This brief is our own summary. Read the full reporting at the source.
